Professional Documents
Culture Documents
Abstract: This paper presents a novel method on such as luminance preserving pseudo-coloring that
coloring the grayscale images. For this purpose, a have been specifically developed to facilitate the
combination of artificial neural networks and some visualization of scientific, industrial, medical and
image processing algorithms was developed to transfer security images [2-4]. Colorization technique is to
colors from a user-selected source image to a target exploits textural information. The work of Welsh et al.
grayscale image. According to the results of this [5], which is inspired by the color transfer and by
combining method, a survey where volunteers were image analogies [6], examines the luminance values in
asked to rate the plausibility of the colorings generated the neighborhood of each pixel in the target image and
automatically was performed for individual images. In adds to its luminance the chromatic information of a
most cases automatically-colored images were rated pixel from a source image with best neighborhoods
either as totally or mostly acceptable. matching.
Keywords: Gray-scale image, neural networks, image In any of those cases, the coloring problem
processing, coloring. amounts to replacing a scalar value stored at each pixel
of a grayscale image (e.g. luminance) by a vector in a
1. Introduction multi-dimensional color space (e.g. a three-
dimensional vector with luminance, saturation and
The color fundamental process followed by the hue). Thus, this is in general a severely under-
human brain in perceiving color is a psychological constrained and ambiguous problem for which it
phenomenon that is not yet fully understood, the makes no sense to try to find an optimum solution, and
physical nature of color can be expressed on a formal for which even the obtainment of reasonable solutions
basis supported by experimental and theoretical requires some combination of strong prior knowledge
results. Basically, the colors we perceive in an object about the scene depicted and decisive human
are determined by the nature of the light reflected from intervention [7]. The color scale selected for
the object. Due to the structure of human eye, all visualization, that is, the sequence of colors used to
colors are seen as variable combinations of the three represent the values of the data range, can have a
so-called Primary colors Red, Green and Blue (RGB). substantial impact on the effectiveness of the
The task of coloring a gray-scale image involves visualization [8].
assigning RGB values to an image which varies along
only the luminance value. Since different colors may In this paper, we present a new methodology to
have the same luminance value but vary in hue or color grayscale images in a fully automatic way that
saturation, the problem of coloring gray-scale images compensates the lack of human intervention by using a
has no correct solution. Due to these ambiguities, database of color images as a source of implicit prior
human interaction usually plays a large role in the knowledge about color statistics in natural images.
coloring process [1]. Adding color to grayscale images Here, we move another step towards automatic
and movies in a way that seems realistic to most coloring of grayscale images, with a methodology
human observers is a problem that greatly challenged where source color images are automatically selected
the motion picture industry in the 1980s and has from an image database. More specifically, we
recently attracted renewed interest within the designed, implemented and experimentally assessed
Computer Graphics community. While coloring of four techniques to choose images from a database
historic photos and classic movies has usually been which was obtained from Fatih University employees
done with the purported intent of increasing their to be used as a source images in color transferring.
visual appeal, there are certain coloring techniques
978-1-4244-3523-4/09/$25.002009
c IEEE 366
2. The Proposed Algorithm
In this section, the general structure of proposed
algorithm describes for transferring color. The general
procedure for converting the gray-scale image to color
image includes two parts: The first part, a digital
image is a two-dimensional image as a set of picture
elements or pixels. Each pixel of an image is
associated to a specific position in 2D region, and has
a value consisting of one or more quantities (samples)
related to that position. The second, Digital images are
classified according to the number and nature of those
quantities:
• binary (bi-level)
• grayscale
• color
• false-color
• multi-spectral
• thematic
Figure 2 RGB shade color wheel and its gray scale Figure 3 Neural Network Architecture
The conversion of RGB image to gray scale image Both training and testing, for each pixel of
minimizes the color space from 16,777,216 grayscale image, 6 values are generated as an input of
(256*256*256) to 256. As a result of minimization of neural network with different methods. As it can be
the color space, many RGB color values are seen in Fig. 4, three different neural networks are used
represented by the same grayscale image value. Hence, to determine red, green and blue components of result
it is impossible to convert grayscale image to RGB RGB image. Three of the network has the same
image using linear conversion methods. To solve this structure which is described above. The network for
non-linearity, neural networks may be used. red component of result RGB image is described as
’network1’, the network for green, is described as
2.2. Coloring the Gray-scale Image Using Neural ’network2’ and the network for blue, is described as
Networks ’network3’. All of these neural networks consist of
Multi Layered Perceptron (MLP) architectures and
In this works, coloring the gray scale image is Back-Propagation training algorithms.
realized with a matlab application using neural
networks. This application is executed in two phases:
y=
(exp( x) + (2 ∗ origin − 1)) (2)
exp( x) + 1
Figure 7 Sample function used for separate data
and
Fig. 8 describes testing phase of structures of ANNs.
3. Application Results
Grayscale images are used as test data. The
grayscale image is input of three networks. When we
combine the outputs of these three networks, we
derivate the output images. Figure 9 shows some test
images and outputs. We notice that the train data is
very important for success. If we use closest images as
train image, performance will be increased. We can
use this technique in coloring old photographs. But we
notice that, there will be some loss in data because of a. test1 input, output and original image
old technology or converting the old photo to digital
medium.